Dataset-Agnostic Recommender Systems

Edoardo D'Amico; Tri Kurniawan Wijaya; Xinyang Shao

REVIEW 3 major objections 3 minor 1 cited by

Dataset-Agnostic Recommender Systems

T0 review · 3 major / 3 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read DAReS proposes that a single zero-configuration codebase, guided only by a small dataset-description file, can replace per-dataset configuration in recommender systems.

desk verdict Position paper with a concrete but under-specified metadata schema; the zero-configuration claim is asserted, not demonstrated, but the proposal is coherent and worth a serious referee. read the letter →

arxiv 2501.07294 v2 pith:AYNGESKX submitted 2025-01-13 cs.IR cs.LG

Tri Kurniawan Wijaya , Edoardo D'Amico , Xinyang Shao This is my paper · ORCID

classification cs.IRcs.LG

keywords dataset-agnosticrecommendersystemsDatasetDescriptionLanguagezero-configurationautomatedfeatureengineeringmodelselectionhyperparametertuningpositionpaperreproducibility

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a new paradigm called DAReS (Dataset-Agnostic Recommender Systems): one codebase that, for a fixed recommendation task, adapts to any new dataset without dataset-specific configuration, feature engineering, or hyperparameter tuning. The load-bearing idea is DsDL, a small structured description of a dataset's columns, types, user/item identifiers, timestamp, and labels, which the system uses to decide preprocessing, model selection, and tuning. The paper is a position paper: it defines the architecture, the metadata grammar, and a roadmap from level-1 (dataset-agnostic, task-specific) to level-2 (task-agnostic) automation. If the claim holds, recommender system development would become reusable and accessible to non-experts, at the cost of reduced per-dataset optimization and higher compute overhead that the paper explicitly accepts.

What carries the argument

The Dataset Description Language (DsDL) is the mechanism that carries the argument. It is a small grammar, formally defined in EBNF, describing a dataset as a list of features with types plus optional user_id, item_id, timestamp, and label fields. DsDL is meant to give the system enough context to treat all datasets uniformly: the type tags drive encoding and preprocessing choices, the presence of user/item columns selects between task families, and the label field determines the prediction target and evaluation metric.

What would settle it

Run the proposed zero-configuration pipeline on a fixed task across datasets that differ in feature types, sparsity, and size, using only a DsDL file as input, and compare against per-dataset tuned baselines. If any dataset produces unusable predictions or requires hidden dataset-specific fixes, the central zero-configuration claim fails.

Watch

Extended reading notes

Core claim

The central claim is that dataset-specific human labor in recommender systems can be replaced by a standardized dataset description. DAReS reads a DsDL file, a YAML/JSON document listing each feature's name and type (categorical, ordinal, numeric, binary, textual, URL), plus optional user ID, item ID, timestamp, and labels, and uses that metadata alone to autonomously perform feature selection, missing-value imputation, feature transformation, noise removal, model selection, hyperparameter tuning, and evaluation. The paper positions this as level-1 automation: dataset-agnostic but still task-specific, with the task (CTR prediction, rating prediction, Top-N) supplied by the user. It argues this eliminates dataset-specific configurations and improves reusability and reproducibility compared to traditional pipelines, while acknowledging a trade-off: less deep customization for specialized datasets and higher computational cost.

Load-bearing premise

The entire proposal rests on the assumption that a short list of column names and types is enough information for an automated system to make all preprocessing, feature-engineering, model-selection, and hyperparameter decisions correctly, without unacceptable performance loss.

Editorial extensions

If this is right

The same DAReS codebase could be applied to a new dataset by writing only a short DsDL file, removing per-dataset code changes.
Non-expert or entry-level users could build recommendation models without domain-specific feature engineering or tuning expertise.
Standardized DsDL descriptions would make experiments easier to reproduce, since dataset treatment is no longer buried in undocumented code.
For highly specialized datasets, DAReS would likely underperform hand-tuned pipelines, and the automated search adds computational cost.
The level-2 extension would let the system infer the task itself from the dataset, removing even the task-specific assumption.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The untested core is whether column-name and type metadata alone is enough to make good model-selection decisions; the most direct way to test that is a fixed task run across many datasets with only DsDL as input.
One could learn the metadata-to-configuration mapping across a corpus of datasets, turning the position paper's proposal into a meta-learning system; the paper does not claim this is done.
The same DsDL schema could be extended with dataset statistics (sparsity, cardinality, skew) that would make the zero-configuration promise more realistic; that extension is not in the paper.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague — this is a position paper, not a systems paper. The reader's take is basically right: the zero-configuration capability claim is asserted, not demonstrated. But I'd push back on the strength of the reject. The paper is honest about being a proposal, it acknowledges the trade-offs in Section 2.5, and the DsDL schema is a concrete first step. If you read it as a vision statement with a proposed metadata standard, it's coherent and readable.

What's actually new: the DsDL grammar (Listing 1) is a small but concrete artifact — six feature types, optional user/item/timestamp, labels. That's not deep, but it's formal enough to implement against. The level-1/level-2 automation distinction is a useful framing for talking about what's achievable now versus later. The writing is clear, and the references to AutoML and reproducibility work are appropriate. I don't see any citation inflation or self-citation problems.

The soft spot is exactly where the stress-test puts it. DsDL records column names and coarse types, but not missingness rates, cardinality, skew, or interaction structure. Two datasets with identical DsDL can easily require different imputation, normalization, and model choices. The paper gives no mechanism, learned mapping, or experiment showing that this metadata is sufficient to drive those decisions. In fact, Sections 2.2 and 2.3 say DAReS 'may' include these components and that a functioning system is possible without them — which reads as the authors themselves not committing to the strong version of the claim. Table 1 admits suboptimal performance. So the abstract's 'autonomously manage' overstates what is actually argued.

That said, this is a position paper, and the absence of experiments is not a fatal flaw if the venue accepts that genre. The paper is a legitimate proposal that could generate useful discussion about metadata standards for recommender systems. I'd send it to review, but with a clear request that the authors either soften the capability claim to match the evidence or provide a proof-of-concept implementation and benchmark. The main value is for researchers working on AutoML for recsys and on reproducibility infrastructure. A serious referee would find the proposal coherent but would need to push on the sufficiency of the metadata.

My recommendation: engage with it as a position paper, not as a claim of results. Peer review can help sharpen the distinction.

Referee Report

3 major / 3 minor

Summary. The paper proposes a new paradigm, Dataset-Agnostic Recommender Systems (DAReS), and a metadata format, Dataset Description Language (DsDL), intended to let a single codebase adapt to arbitrary recommender-system datasets without dataset-specific configuration. DsDL describes column names, coarse feature types, optional user/item/timestamp fields, and labels. The paper claims that this metadata enables autonomous feature engineering, missing-value imputation, feature transformation, noise removal, model selection, and hyperparameter tuning. It also introduces level-1 (dataset-agnostic, task-specific) and level-2 (dataset- and task-agnostic) automation levels. The manuscript is a position paper: it contains no implementation, experiments, benchmarks, or formal correctness arguments.

Significance. If the proposed system worked as claimed, it would be a useful contribution to recommender-system reproducibility, reusability, and accessibility for non-experts. The DsDL grammar is a concrete, citable artifact, and the level-1/level-2 distinction is a reasonable conceptual framing. However, the paper does not establish that DsDL is sufficient to drive the preprocessing, model-selection, and hyperparameter decisions it describes; the core capability claim remains a vision rather than a demonstrated result. There are no machine-checked proofs, reproducible code, or benchmark results to verify the central claim.

major comments (3)

[Abstract, Section 1, and Listing 1] The central claim that DsDL metadata is sufficient for zero-configuration autonomy is unsupported. Listing 1 records only column names and coarse types (categorical, ordinal, numeric, binary, textual, url), plus optional user/item/timestamp/label fields. It does not record missingness rates, cardinalities, skew, distributions, feature interactions, or dataset size, all of which materially affect preprocessing and model choice. Consequently, datasets with identical DsDL descriptions can require different pipelines, and the paper gives no rule or learned mapping from DsDL to the decisions described in Sections 2.2 and 2.3. Section 2.1 itself says the schema may need extension. Without a mechanism or evidence, the abstract's and Section 1's assertions that DAReS eliminates the need for dataset-specific configurations are not established.
[Sections 2.2 and 2.3, Table 1] The manuscript contradicts its own central claim. Sections 2.2 and 2.3 state that autonomous feature engineering, model selection, and hyperparameter tuning are optional, saying it is still possible to create a functioning DAReS without these capabilities, albeit with potentially reduced performance. The abstract and Section 1 promise automatic, zero-configuration operation. Table 1 further concedes suboptimal performance for DAReS due to the generalization-specialization trade-off. The paper needs to either scope the claim to the fully automated configuration and justify that configuration, or weaken the abstract and Section 1 accordingly.
[Section 2.4 and Section 4] No empirical validation, implementation, or falsifiable prediction is provided. The paper describes what DAReS may do and lists evaluation metrics, but it does not run any experiment on diverse datasets, does not provide code, and does not quantify the acknowledged performance and overhead trade-offs. Because the central contribution is an empirical capability claim, a proof-of-concept on even a small set of standard datasets, or a concrete design with testable predictions, is necessary to support it. The future-work statement in Section 4 cannot substitute for this.

minor comments (3)

[Abstract] The abstract contains stray spacing in 'personali zed' and 'signiﬁcan t'; please proofread the abstract.
[Section 2.1] The sentence 'can be derived from other available features and.' appears to be missing a continuation after 'and'; in addition, the EBNF terminal 'String ::= < any string >' is informal and should be defined more precisely if the grammar is meant to be formally checkable.
[Section 3.1] The statement that AutoML approaches 'still require dataset-specific configurations and adaptations' is too broad and is not supported by the cited survey [8]; consider distinguishing DAReS from general AutoML more carefully.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DAReS is a position-paper proposal with no derivation chain; its capability claims are unvalidated rather than circular.

full rationale

The paper contains no equations, fitted parameters, or empirical predictions that reduce to its own inputs. DsDL is introduced as a new metadata schema via an EBNF grammar and a YAML example; it is not derived from any claimed output, nor is any result claimed to follow mathematically from it. The central statements about autonomous feature engineering, model selection, and hyperparameter tuning are phrased as possibilities ('may autonomously handle') and the paper explicitly concedes that a functioning DAReS is possible without these capabilities, albeit with reduced performance. Self-citations appear only as background references (e.g., RBoard, federated learning for CTR prediction, multi-modal collaborative filtering) and are not load-bearing for the paper's proposal. The absence of experiments or external benchmarks is an evidentiary gap, not circular reasoning. Under the provided scoring rubric, this is a non-finding: score 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 2 invented entities

The paper rests on the untested assumption that structured metadata alone can drive all pipeline decisions without performance loss. There are no fitted parameters, but the schema and framework are proposed without independent validation.

assumptions (4)

domain assumption DsDL metadata is sufficient to drive all preprocessing, feature engineering, model selection, and hyperparameter decisions.
Sections 2.2 and 2.3 describe these steps as consequences of DsDL metadata, but no evidence is provided.
domain assumption Zero-configuration adaptation does not incur unacceptable performance loss across datasets.
Section 2.5 and Table 1 acknowledge a trade-off with reduced dataset-specific optimization, but the paper gives no quantification.
domain assumption Existing AutoML and feature engineering techniques can be integrated behind DsDL without dataset-specific reconfiguration.
Sections 2.2 and 2.3 reference external methods such as OpenFE and AutoML, but no integration or test is shown.
ad hoc to paper The DsDL schema is complete enough for all recommender tasks.
Section 2.1 states additional metadata may be needed and will be added later, so completeness is an unsupported working assumption.

invented entities (2)

DAReS (Dataset-Agnostic Recommender System)
purpose: A proposed framework that autonomously handles preprocessing, feature engineering, model selection, and hyperparameter tuning across datasets.
The paper describes the framework but provides no implementation, benchmark, or falsifiable prediction of its performance.
DsDL (Dataset Description Language)
purpose: A proposed standardized metadata schema to make a recommender system codebase dataset-agnostic.
The EBNF grammar in Listing 1 defines the syntax, but there is no implementation, adoption, or benchmark demonstrating that DsDL descriptions enable the claimed automation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Dataset-Agnostic Recommender Systems." pith.science (2026). https://pith.science/paper/AYNGESKX

@misc{pith2026250107294,
  author       = {Pith},
  title        = {Pith review of: Dataset-Agnostic Recommender Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AYNGESKX}},
  note         = {Machine review of arXiv:2501.07294}
}

read the original abstract

Recommender systems have become a cornerstone of personalized user experiences, yet their development typically involves significant manual intervention, including dataset-specific feature engineering, hyperparameter tuning, and configuration. To this end, we introduce a novel paradigm: Dataset-Agnostic Recommender Systems (DAReS) that aims to enable a single codebase to autonomously adapt to various datasets without the need for fine-tuning, for a given recommender system task. Central to this approach is the Dataset Description Language (DsDL), a structured format that provides metadata about the dataset's features and labels, and allow the system to understand dataset's characteristics, allowing it to autonomously manage processes like feature selection, missing values imputation, noise removal, and hyperparameter optimization. By reducing the need for domain-specific expertise and manual adjustments, DAReS offers a more efficient and scalable solution for building recommender systems across diverse application domains. It addresses critical challenges in the field, such as reusability, reproducibility, and accessibility for non-expert users or entry-level researchers.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Universal Reusability in Recommender Systems: The Case for Dataset- and Task-Independent Frameworks
cs.IR 2025-06 conditional novelty 5.0 of 10

The paper proposes a dataset- and task-independent recommender framework (DTIRS) with a schema language (DsDL) and an automation-level roadmap, but provides no implementation or experiments.

Reference graph

Works this paper leans on

29 extracted references · 23 canonical work pages · cited by 1 Pith paper

[1]

Language models are few-shot learners

Tom B Brown. Language models are few-shot learners. arXiv preprint arXiv:2005.14165 , 2020

arXiv 2005
[2]

Secon d workshop on information hetero- geneity and fusion in recommender systems (hetrec2011)

Iván Cantador, Peter Brusilovsky, and Tsvi Kuﬂik. Secon d workshop on information hetero- geneity and fusion in recommender systems (hetrec2011). In Proceedings of the ﬁfth ACM conference on Recommender systems , pages 387–388, 2011

work page 2011
[3]

Item-based top-n r ecommendation algorithms

Mukund Deshpande and George Karypis. Item-based top-n r ecommendation algorithms. ACM Transactions on Information Systems (TOIS) , 22(1):143–177, 2004

work page 2004
[4]

Neu ral architecture search: A survey

Thomas Elsken, Jan Hendrik Metzen, and Frank Hutter. Neu ral architecture search: A survey. Journal of Machine Learning Research , 20(55):1–21, 2019

work page 2019
[5]

Recommenders in the wild - pr actical evaluation methods

Kim Falk and Morten Arngren. Recommenders in the wild - pr actical evaluation methods. In Proceedings of the 17th ACM Conference on Recommender Syste ms, RecSys ’23, page 1, New Y ork, NY , USA, 2023. Association for Computing Machinery

work page 2023
[6]

A trou- bling analysis of reproducibility and progress in recommen der systems research

Maurizio Ferrari Dacrema, Simone Boglio, Paolo Cremone si, and Dietmar Jannach. A trou- bling analysis of reproducibility and progress in recommen der systems research. ACM Trans. Inf. Syst., 39(2), January 2021. 5

work page 2021
[7]

Are we really making much progress? a worrying analysis of recent neural recomme ndation approaches

Maurizio Ferrari Dacrema, Paolo Cremonesi, and Dietmar Jannach. Are we really making much progress? a worrying analysis of recent neural recomme ndation approaches. In Proceed- ings of the 13th ACM Conference on Recommender Systems , RecSys ’19, page 101–109, New Y ork, NY , USA, 2019. Association for Computing Machinery

work page 2019
[8]

Automl: A survey of the state-of-the-art

Xin He, Kaiyong Zhao, and Xiaowen Chu. Automl: A survey of the state-of-the-art. Knowledge-based systems, 212:106622, 2021

2021

Show all 29 references

[9]

The effect of third pa rty implementations on repro- ducibility

Balázs Hidasi and Ádám Tibor Czapp. The effect of third pa rty implementations on repro- ducibility. In Proceedings of the 17th ACM Conference on Recommender Syste ms, pages 272– 282, 2023

2023
[10]

Erase: Benchmarking fe ature selection methods for deep recommender systems

Pengyue Jia, Y ejing Wang, Zhaocheng Du, Xiangyu Zhao, Y ichao Wang, Bo Chen, Wanyu Wang, Huifeng Guo, and Ruiming Tang. Erase: Benchmarking fe ature selection methods for deep recommender systems. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Dat...
[11]

Which is more effective in label noise cleaning, correction or ﬁltering? In Proceedings of the AAAI Conference on Artiﬁcial Intelligence , volume 38, pages 12866–12873, 2024

Gaoxia Jiang, Jia Zhang, Xuefei Bai, Wenjian Wang, and D eyu Meng. Which is more effective in label noise cleaning, correction or ﬁltering? In Proceedings of the AAAI Conference on Artiﬁcial Intelligence , volume 38, pages 12866–12873, 2024

2024
[12]

P osition: A call to action for a human-centered AutoML paradigm

Marius Lindauer, Florian Karl, Anne Klier, Julia Moosb auer, Alexander Tornede, Andreas C Mueller, Frank Hutter, Matthias Feurer, and Bernd Bischl. P osition: A call to action for a human-centered AutoML paradigm. In Ruslan Salakhutdinov, Zico Kolter, Katherine Heller, Adrian ...

2024
[13]

Feature transformation an d subset selection

Huan Liu and Hiroshi Motoda. Feature transformation an d subset selection. IEEE Intell Syst Their Appl, 13(2):26–28, 1998

1998
[14]

Online Meta-Learning for Model Update Aggregation in Federated Learning for Clic k-Through Rate Prediction

Xianghang Liu, Bartłomiej Twardowski, and Tri Kurniaw an Wijaya. Online Meta-Learning for Model Update Aggregation in Federated Learning for Clic k-Through Rate Prediction. 28th ACM SIGKDD 2022 W orkshop on AdKDD, 2022

2022
[15]

Exploring the limits of tr ansfer learning with a uniﬁed text-to-text transformer

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Le e, Sharan Narang, Michael Matena, Y anqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of tr ansfer learning with a uniﬁed text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020

2020
[16]

Intr oduction to recommender systems handbook

Francesco Ricci, Lior Rokach, and Bracha Shapira. Intr oduction to recommender systems handbook. In Recommender systems handbook , pages 1–35. Springer, 2010

2010
[17]

Handling m issing values when applying classiﬁ- cation models

Maytal Saar-Tsechansky and Foster Provost. Handling m issing values when applying classiﬁ- cation models. 2007

2007
[18]

Selecting a classiﬁcation method by c ross-validation

Cullen Schaffer. Selecting a classiﬁcation method by c ross-validation. Machine learning , 13:135–143, 1993

1993
[19]

Generic base standards

Roger S Scowen. Generic base standards. In Proceedings 1993 Software Engineering Stan- dards Symposium, pages 25–34. IEEE, 1993

1993
[20]

RBoard: A Uniﬁed Platform for Reproducible and Reusable Recommender System Benchmarks

Xinyang Shao, Edoardo D’Amico, Gábor Fodor, and Tri Kur niawan Wijaya. RBoard: A Uniﬁed Platform for Reproducible and Reusable Recommender System Benchmarks. CoRR, abs/2409.05526, 2024

2024 arXiv
[21]

Llama: Open and efﬁcient foundation language models

Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timo- thée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, F aisal Azhar, et al. Llama: Open and efﬁcient foundation language models. arXiv preprint arXiv:2302.13971 , 2023

2023 arXiv
[22]

Advancing automation of design decision s in recommender system pipelines

Tobias V ente. Advancing automation of design decision s in recommender system pipelines. In Proceedings of the 17th ACM Conference on Recommender Syste ms, RecSys ’23, page 1355–1360, New Y ork, NY , USA, 2023. Association for Computing Machinery

2023
[23]

Toward e fﬁcient automated feature engineering

Kafeng Wang, Pengyang Wang, and Chengzhong Xu. Toward e fﬁcient automated feature engineering. In 2023 IEEE 39th International Conference on Data Engineerin g (ICDE), pages 1625–1637. IEEE, 2023

2023
[24]

Au- torec: An automated recommender system

Ting-Hsiang Wang, Xia Hu, Haifeng Jin, Qingquan Song, X iaotian Han, and Zirui Liu. Au- torec: An automated recommender system. In Proceedings of the 14th ACM Conference on 6 Recommender Systems , RecSys ’20, page 582–584, New Y ork, NY , USA, 2020. Associat ion for Computing ...

2020
[25]

MM- GEF: Multi-modal representation meet collaborative ﬁlter ing

Hao Wu, Alejandro Ariza-Casabona, Bartlomiej Twardow ski, and Tri Kurniawan Wijaya. MM- GEF: Multi-modal representation meet collaborative ﬁlter ing. CoRR, abs/2308.07222, 2023

2023 arXiv
[26]

On hyperparameter optimization of machine learning algorithms: Theory and practice

Li Y ang and Abdallah Shami. On hyperparameter optimization of machine learning algorithms: Theory and practice. Neurocomputing, 415:295–316, 2020

2020
[27]

Click-through rate predicti on in online advertising: A literature review

Y anwu Y ang and Panyu Zhai. Click-through rate predicti on in online advertising: A literature review. Information Processing & Management , 59(2):102853, 2022

2022
[28]

Openfe: automated feature generation with expert-level performance

Tianping Zhang, Zheyu Aqa Zhang, Zhiyuan Fan, Haoyan Lu o, Fengyuan Liu, Qian Liu, Wei Cao, and Li Jian. Openfe: automated feature generation with expert-level performance. In International Conference on Machine Learning , pages 41880–41901. PMLR, 2023. 7

2023
[2024]

Association for Computing Machinery

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Language models are few-shot learners

Tom B Brown. Language models are few-shot learners. arXiv preprint arXiv:2005.14165 , 2020

arXiv 2005

[2] [2]

Secon d workshop on information hetero- geneity and fusion in recommender systems (hetrec2011)

Iván Cantador, Peter Brusilovsky, and Tsvi Kuﬂik. Secon d workshop on information hetero- geneity and fusion in recommender systems (hetrec2011). In Proceedings of the ﬁfth ACM conference on Recommender systems , pages 387–388, 2011

work page 2011

[3] [3]

Item-based top-n r ecommendation algorithms

Mukund Deshpande and George Karypis. Item-based top-n r ecommendation algorithms. ACM Transactions on Information Systems (TOIS) , 22(1):143–177, 2004

work page 2004

[4] [4]

Neu ral architecture search: A survey

Thomas Elsken, Jan Hendrik Metzen, and Frank Hutter. Neu ral architecture search: A survey. Journal of Machine Learning Research , 20(55):1–21, 2019

work page 2019

[5] [5]

Recommenders in the wild - pr actical evaluation methods

Kim Falk and Morten Arngren. Recommenders in the wild - pr actical evaluation methods. In Proceedings of the 17th ACM Conference on Recommender Syste ms, RecSys ’23, page 1, New Y ork, NY , USA, 2023. Association for Computing Machinery

work page 2023

[6] [6]

A trou- bling analysis of reproducibility and progress in recommen der systems research

Maurizio Ferrari Dacrema, Simone Boglio, Paolo Cremone si, and Dietmar Jannach. A trou- bling analysis of reproducibility and progress in recommen der systems research. ACM Trans. Inf. Syst., 39(2), January 2021. 5

work page 2021

[7] [7]

Are we really making much progress? a worrying analysis of recent neural recomme ndation approaches

Maurizio Ferrari Dacrema, Paolo Cremonesi, and Dietmar Jannach. Are we really making much progress? a worrying analysis of recent neural recomme ndation approaches. In Proceed- ings of the 13th ACM Conference on Recommender Systems , RecSys ’19, page 101–109, New Y ork, NY , USA, 2019. Association for Computing Machinery

work page 2019

[8] [8]

Automl: A survey of the state-of-the-art

Xin He, Kaiyong Zhao, and Xiaowen Chu. Automl: A survey of the state-of-the-art. Knowledge-based systems, 212:106622, 2021

2021